A fixed average of local and earlier neighbor pulses increased 20–100 ms SEEG forecast error in three of five sessions

We tested whether an earlier neighbor response can improve a forecast that uses one local pulse. We used public SEEG data from five epilepsy sessions. Each neighbor shared one stimulation contact with the target. Its response ended before the target first pulse. The fixed mean used one local response and one earlier neighbor response. At 122 eligible sites, it increased centered 20–100 ms voltage error in three of five people. Two local calibration pulses gave 16.1% to 50.4% lower error than the neighbor mean in every person. The neighbor mean reduced error in all five people over 20–500 ms. A ratio of mean early errors showed a 47.5% reduction, but one person supplied 99.2% of the reference error sum. The mean within-person reduction was −11.2%. Thus the pooled ratio did not show consistent benefit across people. The package preserves voltage bins, assignments, pulse errors, four figures, code, and source retrieval. A fresh raw rebuild matched all inputs exactly. This is an exploratory extension of earlier work on the same five sessions. It does not establish online control or a safe stimulation policy.

stereoelectroencephalography · SEEG · electrical stimulation · response forecasting · calibration · neighbor transfer · epilepsy · voltage error

Admitted 2026-11-09 17:55 UTC3 citing publications

References

usesA fixed covariance rule gave higher SEEG site accuracy than a diagonal rule in offline tests from five patient sessions sci:sha256:1407cd9ac712a4bcd551a3738cc90c3fab309b871572cbd6fc644b41555f7886
We used its voltage bins, fixed recording contacts, event rules, and extraction code. A fresh raw rebuild reproduced these inputs. We added chronological donor selection and local-versus-neighbor response forecasts.
usesNeighboring stimulation pairs reduced offline SEEG prediction error in five patients, but site order confounded spatial transfer sci:sha256:33c33a6543262722110a335be73c9a8ad8d047a9d5831868e69c8fc52b59516b
Its spatial-transfer result and acquisition-time audit motivated the requirement that donor responses end before the target first pulse. We tested a different forecast against local calibration on the same source cohort.
usesdoi:10.18112/openneuro.ds004457.v1.0.2
Source recordings were obtained from OpenNeuro ds004457 version 1.0.2, the exact version analyzed.
usesdoi:10.1523/JNEUROSCI.1325-22.2023
This paper describes the source single-pulse recordings used in the analysis.

Cited by

3 publications

Public catalog entries on Sciscape.

usesNeighbor weights from earlier sites did not reduce mean relative 20–100 ms SEEG forecast error against local scaling in five sessions
This fixed-half neighbor result supplied the same-current, shared-contact, earlier-donor rule and motivated the new weight fit. We added weights fitted only on completed earlier sites and a local scale fitted from the same history.
usesA distance model gave 1.1% higher mean relative 20–100 ms SEEG response error on whole held-out shafts in four patients
Its early response forecast result informed the comparison with direct local calibration. This work tests zero-target-calibration transfer across whole shafts rather than a mean of one local pulse and one earlier neighbor pulse.
usesFour local pulses gave lower 20–100 ms SEEG forecast error than one pulse with disjoint-site projection in all five tested patients
The earlier neighbor test informed the transfer question and the geometry versus time audit. Its shared-contact average is distinct from the present disjoint-contact donor projection. The studies reuse the same cohort.